Does automated feedback impact the acceptability of AI-generated police body worn camera review? An implementation science natural experiment

The diffusion of innovations in policing has often been hindered by barriers to implementation and officer acceptance, which can derail an innovation regardless of its validity or effectiveness. Implementation Science (IS) is a useful lens for addressing such concerns because IS empirically examines the way a new technology or strategy is deployed, and it offers insights on barriers, facilitators, and fidelity. The present study uses an IS framework to investigate one of the latest innovations in policing: Artificial Intelligence (AI). In early 2024, two Arizona police departments deployed Truleo, an AI-driven body-worn camera review platform, via randomized controlled trials, but only one of those departments used a feature of Truleo that sends automated positive feedback emails to officers when they engage in behaviors assessed as "highly professional" by the AI algorithm. Using cross-sectional survey data from line-level officers (<i>n</i> = 84), we estimate both intent-to-treat and instrumental variable regression models to examine the effect of the automated emails on three implementation outcomes: acceptability, appropriateness, and feasibility. Officers in the department with automated feedback enabled reported higher levels of the appropriateness of Truleo. The instrumental variable models suggest that the automated emails resulted in higher levels of acceptability and appropriateness. The findings highlight: (1) the importance of evaluating different implementation strategies when deploying new technologies like AI, and (2) the potential value of providing AI-generated positive feedback to officers in the field as a means of ensuring the successful implementation of AI-driven officer accountability platforms in an agency.